The present invention relates to a moving object detection apparatus and a method to correctly extract a moving object from a plurality of input images in a time series.
In the background art, several methods for detecting the moving object from time series images are known. In this case, for example, the moving object is a car moving straightly along a road. In case the moving object and a background are taken by a camera, as methods for detecting the moving object from the input image, the following several methods are considered.
(1) The input image is transformed using motion tracking. The correspondence of a predetermined area between time series images is detected in order to extract only the moving object as the predetermined area. (D. Murray, A. Basu xe2x80x9cMotion Tracking with an Active Cameraxe2x80x9d
(2) A plurality of notice points are located on the image and an area whose peak points are the plurality of notice points is defined. The correspondence of the area between time series images is searched. The affine transformation parameter between the time series image is determined by two steps in order to extract the moving object.
(3) A two-dimensional Laplacian filter is calculated in a moving vector space. The output value is processed by a threshold in order to extract the moving object (Japanese Patent Disclosure (Kokai) PH7-110864).
In the above-mentioned three methods, the background, as the greatest part of the image, is separated from the moving object as another part. However, if the camera is moving along an optical axis direction, movement of the background is not uniformly detected. As a result, the moving object is not correctly detected. Furthermore, while the camera is moving along the optical axis direction, the size of the moving vector of the background is not fixed. Accordingly, a threshold to discriminate a direction or size of a moving vector of the moving object from the image is not determined.
In case the camera moving along the optical axis direction inputs the image including the moving object and the background, the moving vector of the background caused by movement of the camera are generated toward a direction of a vanishing point. Therefore, a part whose moving vector is different from the background is detected as the moving object. However, it often happens that a plurality of edges whose direction is the same as the moving vector of the background are included in the image. A line consisting of the plurality of edges is called a concentration line such as a white line. In this case, the moving vector of the edges is not correctly determined because of an aperture problem.
The xe2x80x9cVanishing pointxe2x80x9d is defined as a point to which a plurality of parallel lines in three-dimensional scene are concentrated in the image in case of the fixed camera. The xe2x80x9cConcentration linexe2x80x9d is defined as the parallel line concentrated to the vanishing point, for example, the white line on the road or the edges of guardrail. In the following explanation, the concentration line is called the white line.
For example, assume that while a car carrying the camera moves along a direction of the white line on the road, the camera inputs the images for a scene behind the car in a time series. The characteristic quantity of the input image includes not only edges of a rear moving car but also the white lines on the road. This white line extends toward the vanishing point generated by movement of the camera car in the time series images. In this case, it is difficult to determine the moving direction of the white line in the image. As a result, the moving object is not correctly extracted from the image.
It is an object of the present invention to provide a moving object detection apparatus and a method to correctly detect the moving object from the image if the camera loaded on the car inputs the rear scene including the moving object and the road.
According to the present invention, there is provided a moving object detection apparatus, comprising: image input means for inputting a plurality of images in a time series; vanishing point detection means for detecting a vanishing point to which a plurality of concentration lines are concentrated from each input image; concentration line elimination means for eliminating the plurality of concentration lines from the each input image according to a position of the vanishing point; and moving object detection means for detecting a moving object from each input image in which the plurality of concentration lines are eliminated.
Further, in accordance with the present invention, there is also provided a moving object detection method comprising the steps of: inputting a plurality of images in a time series; detecting a vanishing point to which a plurality of concentration lines are concentrated from each input image; eliminating the plurality of concentration lines from the each input image according to a position of the vanishing point; and detecting a moving object from each input image in which the plurality of concentration lines are eliminated.
Further in accordance with the present invention, there is also provided a computer readable memory containing computer readable instructions, comprising: instruction means for causing a computer to input a plurality of images in time series; instruction means for causing a computer to detect a vanishing point to which a plurality of concentration lines are concentrated from each input image; instruction means for causing a computer to eliminate the plurality of concentration lines from each input image according to a position of the vanishing point; and instruction means for causing a computer to detect a moving object from each input image in which the plurality of concentration lines are eliminated.